Page 82 - Glucose Monitoring Devices

Why modeling the SMBG measurement error? 81

characteristics and on the other hand to generate synthetic SMBG data that can be
used for the so-called in silico clinical trials. In silico clinical trials (ISCTs) are
defined as “The use of individualized computer simulations in the development or
regulatory evaluation of a medicinal product, medical device, or medical interven-
tion” [21]. The idea is to recreate the concept of in vivo clinical trials (i.e., trial
performed on real patients) in a simulation environment. The strength of ISCT is
that they can overcome some limitations of in vivo clinical trials, such as long dura-
tion, elevated costs and, as a consequence, low numerosity. In fact, given the low
cost and time required to run computer simulations, ISCT can be performed in an
incredibly large number of subjects, that would be impossible to enroll in an
in vivo clinical trial, because too expensive and time consuming. Another advantage
of ISCT is the possibility of testing high-risk situations related to the occurrence of
rare events, not observable in in vivo clinical trials because of their limited size and
duration. ISCTs are thus unique procedures to test the safety of treatments based on
drugs and medical devices under extreme conditions, without exposing human
patients to any risk and can be used to reduce, refine, or even replace in vivo clinical
trials.
Of course, an essential requirement to perform ISCT is the availability of a model
of the patient’s physiological response to the drug or medical device-based treatment
under test, with parameterization that accounts for the interindividual variability and
is, therefore, able to describe a large number of virtual subjects. Diabetes has been an
area of intense modeling development in these last 20 years [22e24]. One of the
most popular models of T1D patients’ physiology is the UVA/Padova T1D simu-
lator, jointly developed by the University of Padova and the University of Virginia
[25e27]. However, a physiology model alone is not sufficient to realize such
ISCT, because, besides the physiological response of T1D subjects to meals and
insulin doses, other fundamental components, such as glucose measurements
collected by glucose monitoring devices, need to be simulated to obtain realistic
in silico scenarios.
The simplest strategy that could be used to generate in silico realizations of
SMBG measurement error is bootstrapping from a dataset of SMBG error realiza-
tions derived from real data. However, a limitation of this approach is that the
number of different SMBG error realizations that can be simulated is finite
(and, in particular, equal to the cardinality of the available dataset). This represents
a problem for large-scale in silico applications, for example, an in silico clinical trial
with long study duration and/or a large number of virtual subjects involved. This
limitation can be overtaken by using a model of SMBG measurement error, which
allows to generate an unlimited number of virtual SMBG error realizations.
Therefore models of the SMBG measurement error allowing the simulation of
synthetic SMBG data are very useful tools to integrate within T1D simulation plat-
forms, like the UVA/Padova T1D simulator, that could then be used to perform
ISCT, for example, to assess the impact of SMBG error on the safety and effective-
ness of SMBG-based treatments.